---
title: Ai Fast Image Server
emoji: 🌍
colorFrom: gray
colorTo: blue
sdk: gradio
sdk_version: 5.47.2
app_file: app.py
license: mit
pinned: false
suggested_hardware: a10g-small
duplicated_from: hysts/SD-XL
load_balancing_strategy: random
---

# AI Fast Image Server

A lightweight Gradio app that serves fast **text-to-image** generation using either:

- **SDXL Base 1.0 + LCM** (default), or  
- **SSD-1B + LCM LoRA** (enable via a flag in `app.py`)

The app targets **very few inference steps** (e.g., 4) for speed while keeping good image quality. It falls back to **CPU** automatically if CUDA isn’t available.

---

## Features

- ⚡ **Fast sampling** with **LCM** schedulers  
- 🔁 **Deterministic** results via seed  
- 🖥️ **Auto GPU/CPU** selection (no brittle `nvidia-smi` checks)  
- 🔐 Optional **secret token** gate to prevent abuse  
- 🧩 Switch between **SDXL** and **SSD-1B+LCM LoRA** with a flag

---

## Requirements

Dependencies are pinned for compatibility (notably `diffusers==0.23.0` + `huggingface_hub==0.14.1`):

```txt
accelerate==0.24.1
diffusers==0.23.0
gradio==3.39.0
huggingface_hub==0.14.1
invisible-watermark==0.2.0
Pillow==10.1.0
torch==2.1.0
transformers==4.35.0
safetensors==0.4.0
numpy>=1.23
ipython